Skip to content

Commit 8e3ff82

Browse files
DavidKorczynskiDavidKorczynski
authored
llamacpp: fix build (#14317)
Signed-off-by: David Korczynski <david@adalogics.com>
1 parent 97728d6 commit 8e3ff82

File tree

5 files changed

+43
-38
lines changed

5 files changed

+43
-38
lines changed

‎projects/llamacpp/Dockerfile‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -17,6 +17,6 @@
1717
FROM gcr.io/oss-fuzz-base/base-builder
1818
RUN apt-get update && apt-get install -y make autoconf automake xxd
1919
RUN git clone https://github.com/ggerganov/llama.cpp
20-
WORKDIR llama.cpp
20+
WORKDIR $SRC/llama.cpp
2121
COPY build.sh $SRC/
22-
COPY fuzzers fuzzers
22+
COPY fuzzers $SRC/llama.cpp/fuzzers

‎projects/llamacpp/build.sh‎

Lines changed: 38 additions & 35 deletions
Original file line numberDiff line numberDiff line change
@@ -16,8 +16,7 @@
1616
################################################################################
1717

1818
export GGML_NO_OPENMP=1
19-
sed -i 's/:= c++/:= ${CXX}/g' ./Makefile
20-
sed -i 's/:= cc/:= ${CC}/g' ./Makefile
19+
2120
# Avoid function that forks + starts instance of gdb.
2221
sed -i 's/ggml_print_backtrace();//g' ./ggml/src/ggml.c
2322

@@ -30,7 +29,12 @@ sed -i 's/ggml_calloc(size_t num, size_t size) {/ggml_calloc(size_t num, size_t
3029
# Patch a potentially unbounded loop that causes timeouts
3130
sed -i 's/ok = ok \&\& (info->n_dims <= GGML_MAX_DIMS);/ok = ok \&\& (info->n_dims <= GGML_MAX_DIMS);\nif (!ok) {fclose(file); gguf_free(ctx); return NULL;}/g' ./ggml/src/ggml.c
3231

33-
UNAME_M=amd642 UNAME_p=amd642 LLAMA_NO_METAL=1 make -j$(nproc) llama-gguf llama-server
32+
# Build with CMake
33+
mkdir build
34+
cd build
35+
cmake .. -DBUILD_SHARED_LIBS=OFF -DGGML_NO_OPENMP=1 -DLLAMA_BUILD_SERVER=ON -DLLAMA_BUILD_EXAMPLES=ON -DLLAMA_BUILD_TOOLS=ON -DLLAMA_CURL=OFF
36+
cmake --build . --config Release -j$(nproc) --target llama-gguf llama-server
37+
cd ..
3438

3539
# Convert models into header files so we can use them for fuzzing.
3640
xxd -i models/ggml-vocab-bert-bge.gguf > model_header_bge.h
@@ -44,32 +48,29 @@ xxd -i models/ggml-vocab-baichuan.gguf > model_header_baichuan.h
4448
xxd -i models/ggml-vocab-deepseek-coder.gguf > model_header_deepseek_coder.h
4549
xxd -i models/ggml-vocab-falcon.gguf > model_header_falcon.h
4650

51+
# Configure flags and libraries
52+
# Note: -lcommon must come before -lllama, and -lllama before -lggml
53+
LIBS="-Lbuild/common -lcommon -Lbuild/src -lllama -Lbuild/ggml/src -lggml -lggml-cpu -lggml-base -Lbuild/vendor/cpp-httplib -lcpp-httplib"
54+
FLAGS="-std=c++17 -Iggml/include -Iggml/src -Iinclude -Isrc -Icommon -Ivendor -I./ -DNDEBUG -DGGML_USE_LLAMAFILE"
4755

48-
mkdir myos
49-
find ./ggml/ -name *.o -exec cp {} myos/ \;
50-
find ./src/ -name *.o -exec cp {} myos/ \;
51-
find ./common/ -name *.o -exec cp {} myos/ \;
52-
OBJ_FILES="myos/*.o"
53-
FLAGS="-std=c++11 -Iggml/include -Iggml/src -Iinclude -Isrc -Icommon -I./ -DNDEBUG -DGGML_USE_LLAMAFILE"
54-
55-
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} fuzzers/fuzz_json_to_grammar.cpp -o $OUT/fuzz_json_to_grammar
56-
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} fuzzers/fuzz_apply_template.cpp -o $OUT/fuzz_apply_template
57-
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} fuzzers/fuzz_grammar.cpp -o $OUT/fuzz_grammar
56+
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} fuzzers/fuzz_json_to_grammar.cpp -o $OUT/fuzz_json_to_grammar $LIBS
57+
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} fuzzers/fuzz_apply_template.cpp -o $OUT/fuzz_apply_template $LIBS
58+
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} fuzzers/fuzz_grammar.cpp -o $OUT/fuzz_grammar $LIBS
5859

59-
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} \
60-
-Wl,--wrap,abort fuzzers/fuzz_load_model.cpp -o $OUT/fuzz_load_model
60+
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} \
61+
-Wl,--wrap,abort fuzzers/fuzz_load_model.cpp -o $OUT/fuzz_load_model $LIBS
6162

62-
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} \
63-
-Wl,--wrap,abort fuzzers/fuzz_inference.cpp -o $OUT/fuzz_inference
63+
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} \
64+
-Wl,--wrap,abort fuzzers/fuzz_inference.cpp -o $OUT/fuzz_inference $LIBS
6465

65-
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} \
66-
-Wl,--wrap,abort fuzzers/fuzz_structured.cpp -o $OUT/fuzz_structured
66+
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} \
67+
-Wl,--wrap,abort fuzzers/fuzz_structured.cpp -o $OUT/fuzz_structured $LIBS
6768

68-
$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} \
69-
-Wl,--wrap,abort fuzzers/fuzz_structurally_created.cpp -o $OUT/fuzz_structurally_created
69+
#$CXX $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} \
70+
# -Wl,--wrap,abort fuzzers/fuzz_structurally_created.cpp -o $OUT/fuzz_structurally_created $LIBS
7071

7172
# Prepare some dicts and seeds
72-
./llama-gguf dummy.gguf w
73+
build/bin/llama-gguf dummy.gguf w
7374
mkdir $SRC/load-model-corpus
7475
mv dummy.gguf $SRC/load-model-corpus/
7576
zip -j $OUT/fuzz_load_model_seed_corpus.zip $SRC/load-model-corpus/*
@@ -87,16 +88,18 @@ cp fuzzers/llama.dict $OUT/fuzz_grammar.dict
8788
cp fuzzers/llama.dict $OUT/fuzz_structured.dict
8889
cp fuzzers/llama.dict $OUT/fuzz_json_to_grammar.dict
8990

90-
if [ "$FUZZING_ENGINE" != "afl" ]
91-
then
92-
$CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} -DFUZZ_BGE fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_bge
93-
$CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} -DFUZZ_BPE fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_bpe
94-
$CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} -DFUZZ_SPM fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_spm
95-
$CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} -DFUZZ_COMMAND_R fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_command_r
96-
$CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} -DFUZZ_AQUILA fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_aquila
97-
$CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} -DFUZZ_QWEN2 fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_qwen2
98-
$CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} -DFUZZ_GPT_2 fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_gpt_2
99-
$CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} -DFUZZ_BAICHUAN fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_baichuan
100-
$CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} -DFUZZ_DEEPSEEK_CODER fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_deepseek_coder
101-
$CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} ${OBJ_FILES} -DFUZZ_FALCON fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_falcon
102-
fi
91+
92+
# Below harnesses are disabled because there seems to be an insta FP in them.
93+
#if [ "$FUZZING_ENGINE" != "afl" ]
94+
#then
95+
# $CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} -DFUZZ_BGE fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_bge $LIBS
96+
# $CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} -DFUZZ_BPE fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_bpe $LIBS
97+
# $CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} -DFUZZ_SPM fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_spm $LIBS
98+
# $CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} -DFUZZ_COMMAND_R fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_command_r $LIBS
99+
# $CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} -DFUZZ_AQUILA fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_aquila $LIBS
100+
# $CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} -DFUZZ_QWEN2 fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_qwen2 $LIBS
101+
# $CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} -DFUZZ_GPT_2 fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_gpt_2 $LIBS
102+
# $CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} -DFUZZ_BAICHUAN fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_baichuan $LIBS
103+
# $CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} -DFUZZ_DEEPSEEK_CODER fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_deepseek_coder $LIBS
104+
# $CXX -Wl,--wrap,abort $LIB_FUZZING_ENGINE $CXXFLAGS ${FLAGS} -DFUZZ_FALCON fuzzers/fuzz_tokenizer.cpp -o $OUT/fuzz_tokenizer_falcon $LIBS
105+
#fi

‎projects/llamacpp/fuzzers/fuzz_apply_template.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -33,7 +33,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
3333
};
3434
size_t message_count = 6;
3535

36-
llama_chat_apply_template(nullptr, p1.c_str(), conversation, message_count,
36+
llama_chat_apply_template(p1.c_str(), conversation, message_count,
3737
true, buf, 4096);
3838
return 0;
3939
}

‎projects/llamacpp/fuzzers/fuzz_json_to_grammar.cpp‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -13,6 +13,7 @@ limitations under the License.
1313
#include "llama.h"
1414
#include "llama-grammar.h"
1515
#include "json-schema-to-grammar.h"
16+
#include <nlohmann/json.hpp>
1617

1718
using json = nlohmann::json;
1819

‎projects/llamacpp/fuzzers/fuzz_structurally_created.cpp‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -11,6 +11,7 @@ limitations under the License.
1111
*/
1212

1313
#include "llama.h"
14+
#include "gguf.h"
1415
#include <fstream>
1516
#include <iostream>
1617
#include <map>

0 commit comments

Comments
 (0)